Problem
Benchmark results are reported as a flat table of the current run. There's no way to see if metrics are trending up or down without manually comparing artifact downloads.
Approach
The benchmark-trend.ts script (from #1866) generates a Markdown trend table appended to $GITHUB_STEP_SUMMARY:
- Latest vs Previous — table showing delta and % change per metric with directional indicators
- Historical table — last 10 runs showing P95 for each metric over time
Example output:
### Latest vs Previous Run
| Metric | Previous P95 | Current P95 | Delta | % Change |
|--------|-------------|-------------|-------|----------|
| container_startup_warm | 18,469ms | 13,200ms | -5,269ms | -28.5% |
### Historical Results (P95)
| Date | Commit | Warm Start | Cold Start | HTTPS Latency | Memory |
|------|--------|------------|------------|---------------|--------|
| 2026-04-09 | 0be647f | 18,469ms | 26,160ms | 70ms | 20MB |
Problem
Benchmark results are reported as a flat table of the current run. There's no way to see if metrics are trending up or down without manually comparing artifact downloads.
Approach
The
benchmark-trend.tsscript (from #1866) generates a Markdown trend table appended to$GITHUB_STEP_SUMMARY:Example output: